Samuel M Bateman

Robotics PhD Student @Princeton | ex-Senior MLE @nuro

[Samuel M Bateman @smbml_]

🚨 Stop feeding your world model the same boring demos. Switch to free-range play data! 🚨

Our CoRL 2026 paper finds we can optimize autonomous data collection to focus on missing action consequences, achieving up to 85x the data scaling efficiency of expert data collection!🧵

[Samuel M Bateman @smbml_]

@BowenFengbf Nice work! I am curious what parts of the reasoning trace degrade the goal seeking behavior?

[Samuel M Bateman @smbml_]

@sundharvs @tennyyin If you try this out, let me know how it goes and if you have any suggestions to make it better!

[Samuel M Bateman @smbml_]

@chichengcc Thanks Cheng!

[Samuel M Bateman @smbml_]

🚨 Stop feeding your world model the same boring demos. Switch to free-range play data! 🚨

Our CoRL 2026 paper finds we can optimize autonomous data collection to focus on missing action consequences, achieving up to 85x the data scaling efficiency of expert data collection!🧵

[Samuel M Bateman @smbml_]

@hazel_heejeong 100%! I saw this and thought it was really topical!

A cool connection: in a deterministic setting, our one-step dataset empowerment objective maximizes reachable goals near the existing dataset.

In other words, we are maximizing the ability to get to AnyGoal!

[Samuel M Bateman @smbml_]

@m_emanaftab The AIM reward should generally favor diverse, action dependent consequences.

If the data already shows knocking an object off a table in diverse ways, repeating those behaviors to get lucky with new outcomes by chance should earn less reward.

Seems fun to try out!

[Samuel M Bateman @smbml_]

5/

Project page: https://t.co/ubs4McwHKr
arXiv: https://t.co/ZR65NrBRkg
Code: coming soon 🐣

This is my first project of my PhD 🚀
Thanks to all my amazing collaborators for the help! @tennyyin @chongyiz1 @Lei_Huang__ @brianmuen @ben_eysenbach @jaime_fisac @shahdhruv_

[Samuel M Bateman @smbml_]

4/

This results in significant data efficiency: 500 actively collected episodes added to 3k demonstrations match a world model trained on 300k expert episodes on end-effector controllability, an 85x reduction in data.

[Lihan Zha @LihanZha]

What’s missing from robot foundation models? An explicit understanding of WHY motions work: why move in this direction, why place the object in this way, how hard to grasp the object. This motion-level reasoning bridges instructions to actions—and provides knowledge that can generalize across embodiments and scenes.

Pure behavior cloning supervises WHAT actions to take, but leaves models to infer WHY those actions work. This becomes especially limiting when learning from large-scale heterogeneous data, including egocentric human data, where humans and robots can take very different actions to achieve the same physical effect. We need models to extract the intent behind those motions, rather than simply recite them. Why leave that knowledge buried in trajectories when we can teach it explicitly?

Introducing EgoLAP: a shared language for human and robot motion. Language actions teach HOW to move; motion-level reasoning teaches WHY the motion works. Together, they unlock, to our knowledge, unprecedented gains in zero-shot performance for robot foundation models without additional robot demonstrations.

https://t.co/S3STiRu42Y
https://t.co/BFHZRXuld5

[Mingtong Zhang @alexzhang_robo]

Introducing our #RSS2026 work! A generator-verifier framework that enables robot policies to improve at inference time, and improve autonomously on-the-job!

🔗 https://t.co/qWAAJoCQcm

[Tenny Yin @tennyyin]

🎮 Can we learn interactive world models from letting robots “play”?

➡️ Introducing ✨PlayWorld: a framework for training high-fidelity video world models from large-scale autonomous play experience that enables:

→ Accurate dynamics prediction
→ Reliable policy evaluation
→ RL fine-tuning entirely inside the world model

🌐https://t.co/Kpd2DoveXc

[Kaiqu Liang @kaiqu_liang]

New Meta Research 🚀

AI agents are powerful, but don’t stay aligned with you over time.

When preferences shift, they don’t adapt. You correct them once…they repeat the mistake. 🤦

Introducing PAHF: continual personalization where agents learn from feedback to stay in sync.

[Jack Morris @jxmnop]

i hate it when the new claude code starts using tools and googling stuff. like that's my job. i google stuff and then paste it to show you what i saw. don't leave me out of this

[Samuel M Bateman @smbml_]

@canondetortugas @max_simchowitz This looks really cool! Are you planning on making a recording available?

[Dhruv Shah @shahdhruv_]

My group @Princeton is hiring!

We are looking for strong postdoc and PhD candidates to join our quest for intelligent robots in open-world environments. Read more below and get in touch 🤖🐅🧡

https://t.co/7o35pwPZCz

[Samuel M Bateman @smbml_]

Super cool work!

[Sunday @sundayrobotics]

Introducing ACT-1: The frontier robot AI behind Memo.

ACT-1 enables Memo to perform: 🎭
⏰ Lengthy tasks like cleaning up after your dinner
🦾 Dexterous tasks such as folding socks
☕ Daily delights like making an espresso

How does it work? 🧵

[Haimin Hu @HaiminHu]

Grew up dreaming of reliable robot allies from movies and anime? 🤖

Join the Alliance AI Lab @JHUCompSci @HopkinsDSAI to turn your childhood imagination into reality as we advance the algorithmic foundations of human-centered robotics.

PhD: apply by Dec 15.
Intern: email me!